Looks like Nehalem is a monster in SPECint and SPECfp.


To those who don't understand the graph, ie ZOLDMAN, the line graph represents Intel's projections of Opteron performance. The last point on the graph labeled SH 2.8 GHZ, represent Intel's projections for Shanghai 2.8 GHZ. The two bars that says BAR 2.3 GHZ represents Barcelona.

Obviously SPEC scores don't always correspond to desktop performance. But we do know that Core2 is already head and shoulders above Barcelona. The only question for Nehalem left to ask is how much. We know that Nehalem will IMC and will have 3 channel DDR3 vs 2 for Barcelona. Quickpath is also a direct connect architecture like Hypertransport so AMD's advantage in scaling will be negated. From what I've heard Quickpath is suppose to be even better than HT3.

That said, Barcelonas still don't exist on the market. At this rate, Intel might launch Nehalem before Barcelona.

I dont understand the graph

ditto, and, where's thunderman?

The only thing we need to understand is "PROJECTIONS".

Fair is fair. We hammered AMD for "simulated" benchmarks based on a nonexistent processor.

yup, its allllllllll bullcrap

more intel optimised compiling I see ....

jsc is right.

I dont get it.....also 60%?? I find that hard to belive, even if it is a REALLY god architecture etc, 60% thats a bit of a fudged figure i think.....

Remember, nehalem is supposed to scale to 8 cores, so 60% doesn't sound that outlandish, and since Intel has been on a streak as of late, I'm inclined to believe it...

It's spec_*_RATE.

What you are mainly seeing is the QPI (FSB replacement) in action.

You can extract nothing regarding raw single-threaded clock-for-clock performance.

(oh, an I agree -- these are Intel marketing slides... so you can also extract nothing *real* from these plots. Just trying say that the memory bandwidth improvement is what could be responsible for intels *projected* gains)

The chart is also making its assumption based on the failed Barcelona/Phenom series. If the B3 Barcelona/Phenom performs better, and most people expect that it will, then the better performance margin could drop dramatically. Not saying that it will, only that it could.

That said, the chart is BS and deserves to get hammered just as the AMD did for "simulated" benchmarks, just as jsc wrote.

Hmm, I really don't doubt it. I mean, when considered, most of the Core 2 line now is already bottlenecked by the FSB. The highest ones go up to 1600 MHz. However, Nehalem is much faster, with the highest processors reaching out at twice that. On top of that, every core will have a hyperthread, so even the low end dual-cores will have 4 threads. Add in the remade caches, the added instructions, and the overall reworked architecture ... I find it very plausible.

I mean, clock for clock, a Core 2 Quad already beat out a Phenom by about 10%. Nehalem is supposed to have up to 25% increased on single threaded apps, and up to a full 100% increase on multi-threaded apps. From the demonstration shown some time ago, it all seems like it's going to work out like that.

Almost.
The chart is based on 2006 CPU's and the boost of the newer Intel upgrades...with no mention of an Optron upgrade at all (Intel CPU #'s are noted).
We would expect to see a boost in newer CPU's.

The chart -also- notes "HPC" systems and fails to state how many CPU's they are useing in the test.
Cray HPC's no longer use Intel nor do any of the HPC's used by the movie industry...or the world largest HPC Super Computers.

HPC's are not desktops nor are they 2-4 CPU socket systems used as servers in data centers,yet the claim of 60% more seems they are useing very few CPU's (perhaps even only two) to "fudge" the graph further.
After reaching X number of CPU's the dirrect CPU interconnect system AMD uses runs rings around a chipset controled Intel system.

EDIT: Looking closer at the chart the so called HPC system is in fact -only- a two CPU system...not a true HPC.
MORE EDIT: What dual core 3Ghz Optron are the useing for this test? I know of no such CPU.
EVEN MORE EDIT: The longer I look at the chart the worse it looks. Intel seems to have been bested by slower Optrons in the first three results and the last result they decided to use a 2.8 Optron to test against.

Intel has always been killed in these tests... they are largely a function of memory bandwidth. This is the place where AMD held (and continues to hold) the ground due to their integrated IMC/ memory bandwidth advantage.

Forget about the raw numbers and strange comparisons. The take-home message is this -- intel is telling their vendors that they expect that their QPI solution will erase AMDs memory throughput advantage. Whether QPI performs as well as intel is claiming has yet to be confirmed.

What's it say?
Nehalem will have almost double the fpu and int perf of current xeons.
Since Spec is super optimized for core2 (the L1 cache miss rate is 1/3 of best available in the real world), and only uses ~ 25% of processor power, it is not my idea of a great benchmark. Just the same, it looks likw nehalem will be great.